-
B.J. Allen, Deepa Chandrasekaran, & Suman Basuroy
Design Crowdsourcing: The Impacton New Product Performance
ofSourcing Design Solutions from
the “Crowd”The authors examine an increasingly popular open
innovation practice, “design crowdsourcing,” wherein firms
seekexternal inputs in the form of functional design solutions for
new product development from the “crowd.” They
investigateconditions under which managers crowdsource design and
determine whether such decisions subsequently boostproduct sales.
The empirical analysis is guided by qualitative insights gathered
from executive interviews. The authorsuse a novel data set from a
pioneering crowdsourcing firm and find that three concept design
characteristics—perceivedusability, reliability, and technical
complexity—are associated with the decision to crowdsource design.
They use aninstrumental variable method accounting for the
endogenous nature of crowdsourcing decisions to understand whensuch
a decision affects downstream sales. The authors find that design
crowdsourcing is positively related to unit salesand that this
effect is moderated by the idea quality of the initial product
concept. Using a change-score analysis ofconsumer ratings, they
find that design crowdsourcing enhances perceived reliability and
usability. They discuss thestrategic implications of involving the
crowd, beyond ideation, in helping transform ideas into effective
products.
Keywords: product design, crowdsourcing, user design, open
innovation, new product development
Online Supplement: http://dx.doi.org/10.1509/jm.15.0481
Increasingly, firms are tapping into a wide range of
externalsources of knowledge to source innovations (Chesbrough2003;
Laursen and Salter 2006). One popular facet of thisnew trend is the
leveraging of online infrastructure to tap anunderexplored and
richly heterogeneous pool of knowledgeresident in the general
population of consumers for innovativeideas (Bayus 2013), a
practice termed “crowdsourcing.” Extantresearch on the efficacy of
using external sources of knowledgefor innovation has centered on
the opportunity identificationstage (Foss, Lyngsie, and Zahra
2013). For example, research inmarketing has examined the practice
of involving the crowdduring ideation (e.g., Bayus 2013; Poetz and
Schreier 2012) andhas suggested that such early involvement in new
productdevelopment (NPD) empowers potential consumers while
enabling firms to attract more participants overall and
morediverse participants to the idea generation process
(Fuchs,Prandelli, and Schreier 2010; Schreier, Fuchs, and Dahl
2012).
Recent evidence has suggested that the role of externalknowledge
sources may go beyond opportunity identificationand extend to
opportunity exploitation stages (Foss, Lyngsie,andZahra
2013).Congruentwith this idea,firms are increasinglyusing
crowdsourcing in phases following ideation, specifically,in the
solicitation of actionable design solutions (see Table 1),
apractice we call “design crowdsourcing.” We define
designcrowdsourcing as the practice of soliciting functional
design1solutions from the crowd. For example, crowdsourcing
plat-forms, such as Redclay.com, allow firms to submit newproduct
design briefs and seek crowd input for the develop-ment and/or
refinement of the design. In such situations, theclient firm
already has a new product idea, but may lack theresources or
know-how to bring this idea into fruition, and,therefore, seeks
external input from the broader user com-munity to help create a
manufacturable design.
B.J. Allen is Assistant Professor of Marketing, Sam M. Walton
College ofBusiness, University of Arkansas (email:
[email protected]). DeepaChandrasekaran (corresponding author)
is Assistant Professor of Mar-keting, University of Texas at San
Antonio (email: [email protected]). Suman Basuroy is
Department Chair and Graham WestonEndowed Professor of Marketing,
University of Texas at San Antonio (email:[email protected]).
The authors thank the Carolan ResearchInstitute and Dr. Joel
Saegert for helping fund this research; participants atthe
Marketing Science Conference and PDMA Research Forum for
theirhelpful comments; and Dr. Ram Ranganathan, Dr. Raji
Srinivasan, andDr. Richard Gretz for detailed comments on earlier
versions of the article.The authors are deeply grateful to all
interviewees for their invaluableinsights. Michael Haenlein served
as area editor for this article.
1Many of the design crowdsourcing platforms focus on
themanufacturing makeup of design; thus, our study focuses on
func-tional design rather than aesthetics. This use is also
consistent withextant research studying the early phases of design
with a “focus onfunctional performance in product design, as
opposed to the product’saesthetic qualities or appearance” (Dahl,
Chattopadhyay, and Gorn1999, p. 19).
© 2018, American Marketing Association Journal of MarketingISSN:
0022-2429 (print) Vol. 82 (March 2018), 106–123
1547-7185 (electronic) DOI: 10.1509/jm.15.0481106
http://dx.doi.org/10.1509/jm.15.0481http://Redclay.commailto:[email protected]:[email protected]:[email protected]:[email protected]://dx.doi.org/10.1509/jm.15.0481
-
TABLE 1Examples of Companies Utilizing Design Crowdsourcing and
Design Crowdsourcing Platforms
Organization Company Description Use of Design Crowdsourcing
Quirkya Pioneering socially developed product companyfounded
with the vision of making invention accessible.
After selecting a new product idea, Quirky asks itscommunity to
help with the design by “[submitting]sketches, images, videos, and
prototypes that illustrateindustrial design directions for [the
product].”
Crowdspringa Platform for bringing companies and
designerstogether
Various companies post creative briefs with the needsand
requirements for new industrial and consumerproducts. Designers
submit design concepts for theproduct and the firm chooses the ones
to utilize.
Unilever Consumer packaged goods company Unilever operates an
open-innovation website,Unilever Foundry, where it collaborates
with itscommunity, and many projects involve designcrowdsourcing.
As stated on its website, “Often we willhave specific challenges on
which we’d welcome yourcollaboration: a new formula, a new
technique, newpackaging or a fresh design solution to a product
wealready have in mind ” (Unilever 2016, emphasisadded).
Fiat Global car company Fiat crowdsourced the design of its Fiat
Mio: “Fiat, soughta design for its 2009 concept vehicle, the Fiat
Mio. Ratherthan turning inward to its core team of designers
andengineers to come up with the new look, the company...let the
world decide how the car would look, feel, anddrive” (Markowitz
2011).
eYeka Third-party website that serves as crowdsourcingplatform
for brands. Client list includes Procter &Gamble, Nestlé, and
Citroen
Firms post ideas for new products along with creativebriefs and
ask the community to submit design ideas.For example, one firm
asked for designs for a newinteractive learning and entertainment
product forchildren, and the community submitted design ideas.
Local Motors Open-innovation car company and community Local
Motors launches a car idea to its community andasks for help in
creating and implementing the design.The community, consisting of
designers andengineers, collaborates and submits designs for
thevarious car parts.
GeneralElectric(GE) OpenInnovation
Branch of GE that generates ideas from consumersvia
crowdsourcing challenges
GE gives a description of the product it is looking for,along
with a few sample sketches, and asks thecommunity to submit design
concepts. Submissionsinclude text descriptions of how the product
works,along with pictures/sketches. Winners receive
cashrewards.
Hyve Crowd A German third-party crowdsourcing site.
Clientsinclude Audi and BMW.
Firms post various requests for product-related ideas,many of
which include design requests. For example,companies highlight a
specific type of product they arelooking for and ask the community
to submit conceptsand design solutions.
Red Clay Platform that connects brands with a community
ofindustrial designers.
Small businesses submit product design briefs that areworked on
by a community of industrial designers. Theproject is matched to a
small group of designers whosubmit designs that are then chosen by
the firm ina contest format. Thereafter, the firm owns all IP.
Design2Gather Third-party crowdsourcing platform designed to
helpfirms develop ideas into actionable designs
Companies post product ideas that are then worked onby hundreds
of designers to develop a manufactureready design. Their tagline is
“making your ideareality.”
aSources of the data for the empirical analysis in the
study.
Design Crowdsourcing / 107
-
These compelling anecdotal examples raise interesting re-search
questions: Does design crowdsourcing lead to a betternew product
development process? Does design crowdsourcinglead to improved
products and performance? Whether and howcrowdsourcing affects
critical downstream activities, such asexecuting ideas in the NPD
process, has received little researchattention. In fact, studies
have highlighted the significant chal-lenges of internalizing
external input into the NPD process, in-cluding the rejection of
outside input by insiders (Katz and Allen1982), the costs of
distant searches (Afuah and Tucci 2012), andthe difficulty of
communicating tacit information needed forproblem solving (Von
Hippel 1994). Furthermore, scant researchhas linked crowdsourcing
to product performance (for an excep-tion, see Nishikawa, Schreier,
and Ogawa 2013), and current lit-erature has focused primarily on
crowdsourcing during ideation.
The goals of this research are to examine (1) whether andhow
design crowdsourcing affects the NPD process; (2) whatdesign
antecedents lie behind the decision to crowdsource; (3)whether
design crowdsourcing has a positive impact on
productperformance—and, if so, to identify some boundary
conditionsfor this effect; and (4) whether design crowdsourcing
helpsimprove the functional design attributes of product ideas.
Weuse the knowledge-based theory (Alavi and Leidner 2001;Chang and
Taylor 2016), as well as exploratory insights frominterviews
(suggested by Kumar et al. [2016] to uncover newphenomena), to
propose that (1) the crowd is a repository ofdesign knowledge and
design crowdsourcing is a mechanismthat enables firms both to tap
into the broader community forworkable design solutions and to
assimilate/exploit these so-lutions to aid the transformation of
new product ideas intoproducts, (2) such identification and the
exploitation of externaldesign solutionswill improve newproduct
performance, and (3)the efficacy of design crowdsourcing on
performance willdepend on the quality of the original product
ideas.
We test our hypotheses using a novel data set of 86 newproducts
collected fromQuirky, a pioneering, community-drivenNPDwebsite. The
empirical analysis on this data set indicates thatthe probability
of design crowdsourcingwas influenced by a needto increase
perceived usability and reliability and to decreasetechnical
complexity. Using an instrumental variable procedure(Wooldridge
2010) for dealingwith endogenous binary variables,we find that
design crowdsourcing has a positive effect on sales,as proposed,
with an important boundary condition. The positiveimpact of design
crowdsourcing on sales is contingent on the ideaquality of the
original product concept—design crowdsourcing isassociated with
increased sales when the idea quality of theproduct concept is low.
Furthermore, design crowdsourcingenhances perceived reliability and
usability from idea to finalproduct.
Our results suggest that design crowdsourcing can helpmanagers
move a greater number of ideas through developmentby using the
community’s assistance in making (initially) less-promising
ideasmarketable, thus improving the effectiveness ofthe NPD
process. Rather than discarding such ideas, firms mayuse external
sources of knowledge to develop them and interactwith these sources
extensively to ensure that the outcome is ofhigh quality. In
addition, we highlight specific design function-alities that
managers can improve using design crowdsourcing,allowing for a more
targeted approach when leveraging
crowdsourcing. Finally, our findings suggest opportunitiesfor
crowdsourcing platforms tomarket themselves as solutionspaces that
provide tangible downstream benefits throughenhanced functional
attributes. The next sections present theconceptual development as
well as managerial insights intodesign crowdsourcing leading to the
hypotheses, data, mod-eling methodology, results, and
discussion.
Conceptual DevelopmentDesign Crowdsourcing and
KnowledgeManagement
Grant (1996, p. 112) states that “fundamental to a
knowledge-based theory of the firm is the assumption that the
critical inputin production and the primary source of value is
knowledge.”Afirm’s ability both to create new knowledge and to
applyknowledge forms the basis of developing a competitive
ad-vantage (Alavi and Leidner 2001). One online mechanism
thatgrants firms access to a wide, diverse knowledge pool
iscrowdsourcing (Schreier, Fuchs, and Dahl 2012). Extant mar-keting
literature has treated the crowd as a resource base for newideas
and treated crowdsourcing as amechanism that enables
theidentification of new ideas from this resource base.
However,firms may also need to engage external resources, such as
thecrowd, for opportunity exploitation (Foss, Lyngsie, and
Zahra2013).Wepropose that the crowd is also a knowledge source
fordesign solutions, which are utilized to solve
firm-specificproblems in the context of new product
development.
Knowledgemanagement theory suggests that the
identification/acquisition and assimilation/exploitation of
externally gen-erated knowledge improves innovation performance
(Cohenand Levinthal 1990). However, in the context of
designcrowdsourcing, little is understood about how this
processmanifests itself in practice. Given the lack of research
intodesign crowdsourcing, our research begins with
qualitativeinterviews to investigate this question and then
integrates thefindings from the interviews with a literature
review.
Qualitative Interviews
We conducted exploratory interviews with practitioners fromthe
United States, China, Italy, Israel, and the United Kingdomwho had
extensive experience with crowdsourcing (seeTable 2). Because the
purpose of these interviews was to assistin theory development, we
ensured that the interviewees werefamiliar either with the
experiences of established firms thatengaged in crowdsourcing or
with start-ups whose businessmodels involved crowdsourcing. We
followed a standardformat and approach for each interview.2 The
authors carefully
2After a brief description of the research project, each
intervieweewas asked about issues related to crowdsourcing and how
thoseoutside the organization help with providing design solutions.
Inthree of the cases, the interviewee chose to reply to these
questionsby email, in which case further emails were sent to follow
up onresponses, if needed. We supplemented these insights with a
searchfor popular press articles to gain a broader understanding of
howcrowdsourcing aids product development, using the
LexisNexisdatabase, as well as by examining firms’ internal
websites, designcrowdsourcing websites, and blog posts.
108 / Journal of Marketing, March 2018
-
read the interview transcripts and notes and documented themain
concepts and themes that emerged.
Insights on Design Crowdsourcing and NewProduct Performance
In this subsection, we explore the link between design
crowd-sourcing and new product performance by utilizing
commonthemes from the interviews and the literature review.
Design crowdsourcing helps firms move product ideasinto
development. “How can I execute my innovative ideas?”is a question
that represents an increasing concern for chiefexecutive officers
and business executives (eYeka 2016). Forexample, an executive of a
design crowdsourcing firm notedthis about her clients:
These people come with new ideas in innovation; they have agreat
innovation, but they’re not really sure how to make thatinnovation
happen. (Cofounder and chief operating officer,design crowdsourcing
firm)
Design crowdsourcing helps make development a realityin
situations where firms know what product or solution theywant but
are looking for an executable design. The differencebetween using
the crowd to obtain design solutions and usingthe crowd for
ideation itself seems to be twofold: First, theemphasis of ideation
crowdsourcingmay be on an unconstrainedflow of ideas, whereas
design crowdsourcing involves the crowd
tackling a focused need and, thus, all submissions and
iter-ations are working toward solving the same problem. Second,the
solution space in design crowdsourcing may also besmaller (i.e.,
more manageable). From our investigation ofdesign crowdsourcing
websites, the number of design sub-missions (being in the tens or
hundreds and not thousands, as,for instance, in ideation
challenges) were more tractable forclients (especially small
businesses). Thus, design crowd-sourcing moves product ideas closer
to development and,thus, to delivering value:
Different crowds [are viewed] as layers of technology that
canpowerfully work together. In online sourcing from a crowd
...you’re connecting multiple people to get to an end solution.The
ideas become more powerful when you bring thesedifferent skill sets
together ... to get to a solution a little moreefficiently.
(Cofounder and chief operating officer, third-partydesign
crowdsourcing platform)
Design crowdsourcing helps identify new sources for andtypes of
design solutions. A consistent theme from our in-terviews was that
although in-house specialists may be con-strained by their past
experience while trying to create newdesign solutions,
crowdsourcing brings in novel and freshsolutions to design
problems. For example, when asked whymanagers would crowdsource
product design, one expert noted:
[Managers are influenced by] a desire to bring a fresh insight
tothe design process. Crowdsourcing can help with the design
TABLE 2Description of Managerial Interviews
No. Title Firm Description
1 Vice President, Marketing and Technology Consulting firm,
helps organizations with crowdsourcing &marketing
2 Co-Founder and Chief Operating Officer Third-party design
crowdsourcing platform
3 Senior Brand Manager Large manufacturing firm known for use of
crowdsourcingproduct design
4 Editor (researches and publishes articles oncrowdsourcing)
Website for business news, research, and insights
5 Brand Manager Large CPG firm that organizes numerous
crowdsourcingcampaigns
6 Consumer Trends Consultant (consults on crowdsourcing)
Marketing consulting firm that advises organizations onconsumer
trends
7 Founder and Chief Financial Officer European crowdsourcing
company
8 Content Manager European crowdsourcing company
9 Founder Consulting firm that helps firms facilitate open
innovationchallenges
10 Creative Director Third-party crowdsourcing firm that works
exclusively indesign crowdsourcing
11 Co-Founder and Chief Operating Officer Third-party design
crowdsourcing site with emphasis on“design challenges”
12 Senior Technologist (led development of first crowd-sourced
laptop)
One of the largest computer manufacturers in the world
13 Former President One of the largest crowdsourcing firms in
the world
Design Crowdsourcing / 109
-
process by bringing new ways of thinking and unique ideas.An
in-house team can be impacted by things like legacy ideas,office
politics, and being too close to the product. By bringingin outside
help it brings a fresh approach, which aids thecreative process.
(Editor and author on crowdsourcing)
When asked whether there were differences in the kinds
ofsolutions companies were looking for, an executive commentedon
the criticality of diversity of perspectives:
I think it really depends on the company because some of
oursmaller and medium sized companies don’t have any designtalent
in-house, so they’re really looking for that design. Thenthe
companies that do have design talent in-house, they’relooking to
get more of a new perspective and understandingthat when you pull
more than two designers into a project,you’re going to get a very
diverse amount of perspective,which starts to really begin the true
design thinking of whywedesign and go through the full process,
which is pulling thosedifferent ideas together, iterating on them
this idea thatthere’s a community to build on them versus one
person’s wayof thinking. (Cofounder and chief operating officer,
designcrowdsourcing firm)
This point is consistent with the literature that finds that
userinvolvement in design generates greater numbers of
diverse,need-specific, and unconstrained designs (Schreier,
Fuchs,and Dahl 2012) compared with in-house design.
Furthermore,managers believed that the utilization of the crowd led
to agreater congruency between design and user needs. As notedby a
leading design expert:
Just having ideas doesn’t work. The question is, really, who
areyou solving it for? Insights and the human side of design is
themost important aspect you can add.... I think using design as
adifferentiator is what we’re seeing in the market. (Former
pres-ident of a leading crowdsourcing firm/design consultancy
group)
Managers are continually looking for ways to respond
toconsumers’ wants and needs in a way that optimizes firm
re-sources (Fennell and Saegert 2004). Firms are thus able to
usedesign crowdsourcing to integrate knowledge to develop aproduct
more congruent with consumer needs, which is morelikely to succeed
when it enters the marketplace.
Design crowdsourcing increases available resources forNPD.
Nearly every manager interviewed mentioned that de-sign
crowdsourcing serves as a resource-supplement strategythat
simplifies and accelerates the flow of the NPD process:
A lot of those (client) companies are small. They need
tomovequick and they need to keep their prices down, so
budgetbecomes a big concern. Innovation becomes a
concern.(Cofounder and chief operating officer, design
crowdsourcingfirm)
I’m doing all the things that I’m doing as a typical
productdevelopment cycle, but I’m actually accelerating that
bygetting the crowd involved.... You know your product, youknow
your design. You knowwhat you’re good at, but you’reintentionally
leveraging the crowd to get into the market fast.(Vice president,
crowdsourcing consulting firm)
Popular press publications also use words like
“efficiency,”“simplify,” and “streamline” when describing why firms
crowd-source during NPD. Traditional NPD processes are
constrainedby resource availability, such that only a small number
ofthe “best” ideas can be implemented. Accessing the crowd
increases the knowledge resources available to a firm byboth
leveraging the skills and expertise of hundreds of peopleoutside
the organization and freeing up firm resources, allowingfor the
development of a greater number of ideas.
Design crowdsourcing may be iterative and collab-orative. Design
crowdsourcing is not just about obtainingnew ideas but also about
refining and fine-tuning ideas, and itprovides the capability to
engage in a high degree of collab-oration with the broader
community. This represents one ofthe key differences from
traditional ideation crowdsourcing,wherein the firm may select
novel ideas, but there is not muchcollaboration going forward
(Bayus 2013). The selected de-signers, suppliers, and clients often
(depending on the crowd-sourcing platform)work together
collectively, using insights theygain from design submissions to
iterate toward a manufacture-ready product. Because members of the
“crowd” are neitherfamiliar specialists nor a part of the internal
team, there is a needfor closer monitoring and internal involvement
to move toward asolution. Furthermore, the process of iteration
often results in abetter translation of tacit suggestions to
workable solutions:
My crowd is going to be an extended team within mycompany. (Vice
president, crowdsourcing consulting firm)
Today we have an on demand industrial design community....They
start to look at a lot of these different crowds as layers
oftechnology that can powerfully be able to start the work
to-gether. In online sourcing from a crowd [you are not just]
ableto connect [with one person], but you’re connecting
multiplepeople to get to an end solution. (Cofounder and chief
op-erating officer, design crowdsourcing platform)
In summary, a firm has the choice of whether to involve thecrowd
in the design phase or to simply refine the product designin-house.
Our exploratory insights suggest that design crowd-sourcing enables
firms to (1) translate ideas into executable so-lutions, (2)
provide access to new sources that can provide noveland meaningful
design solutions, (3) help increase the resourcesavailable for NPD,
and (4) help create a more iterative andcollaborative process in
integrating external solutions with in-house guidance. Because a
firm’s “ability to identify, assimilate,and exploit knowledge from
the environment” is related to afirm’sinnovative performance (Cohen
and Levinthal 1990, p. 128), itfollows that identifying,
assimilating, and exploiting knowledgeusing design crowdsourcing
should increase a new product’sperformance.
In the specific case of NPD, we expect all four of thesefactors
to contribute to new product success, as prior literaturehas
suggested that (1) the development of an increased numberof new
products that more accurately reflect customer prefer-ences during
the NPD process improves NPD performance(Joshi and Sharma 2004);
(2) design newness and creativesolutions that are novel and
meaningful to consumer needs arekey determinants of new product
success (Talke et al. 2009); (3)slack creates resources that help
better exploit existing com-petencies, explore new competencies,
and develop innovations(Atuahene-Gima 2005); and (4) conscious and
meaningfulcustomer interactions in the form of engagingwith the
design ofnew products (along with or in addition to ideas) will
provide adifferentiating advantage to the firm in themarketplace
(Ramaniand Kumar 2008).
110 / Journal of Marketing, March 2018
-
Thus, drawing on the insights derived from our interviewsand
past theory, we propose:
H1: Design crowdsourcing has a positive effect on new
productperformance.
Moderating Effect of Initial Idea Quality
A key premise of the prior hypothesizing on
crowdsourcing’spositive effect on new product performance is that
crowd-sourcing the design will help make the product more
market-able. What if the initial raw concept (idea) was
alreadymarketable? Kornish and Ulrich (2014) establish that
betterideas, as assessed by commercial value (purchase intent of
theraw concepts), lead to increased sales. Their finding raises
twoimportant questions: (1) Is there incremental value added
byinvolving the crowd in suggesting design solutions if the
initialproduct idea itself is good? and (2) Can firms extract value
fromlower-quality ideas rather than from discarding them?
The managerial insights showed that design crowdsourcinglikely
leads to an evolutionary process of the new product idea.As one
manager said of a product that she managed, “Theproduct just kept
developing and iterating.” Because designcrowdsourcing draws on the
knowledge of the crowd to im-prove functional attributes and
involves a process of iteration, itis likely that product ideas
with a significant need for im-provement will benefit most from the
process. We propose thatwhen the idea quality is low, the
incremental value of designcrowdsourcing will be high. When the
initial quality of the rawidea is high, the firm might be better
off with in-house design.
This line of thinking is consistent with the broader nature
oforganizational conflict in the exploration of new and
exploi-tation of current knowledge. Andriopoulos and Lewis
(2009)conduct a comparative case study approach of five
leadingambidextrous firms in the product design industry. They
notethat whereas exploitation demands efficiency and
convergentthinking to improve product offerings, exploration
involvessearch and experimentation efforts to generate novel
re-combinations of knowledge, creating tension. Furthermore,there
is tension between, on the one hand, the use of stan-dardized best
practices for NPD that may breed rigidity, and, onthe other hand,
engagement in new routines that may bring infresh thinking and free
up resources but may also be less ef-ficient. Organizations often
have best practices and routines inplace to progress their most
promising ideas with their in-houseresearch and development/design
teams. Thus, for the bestideas, design crowdsourcingmay be less
beneficial, because thechallenges associated with processing and
assimilating new anddiverse design solutions may outweigh potential
benefits.However, for less marketable ideas, design crowdsourcing
mayfacilitate the process with better interaction to help evolve
anddevelop ideas, leading to better performance.
Furthermore,lower-quality concepts can be used as an opportunity to
learnwhich attributes are important to customers, which in turn
helpsfirms develop higher-quality products (e.g., Ries 2011, p.
107).Design crowdsourcing can help uncover such attributes
toimprove the NPD process. Thus, we propose,
H2: The positive impact of design crowdsourcing on new
productperformance is greatest for products with low initial
ideaquality.
Functional Design Attributes as Antecedents toDesign
Crowdsourcing
Design crowdsourcing is not a one-size-fits-all strategy to
beleveraged ubiquitously. Rather, as one executive noted, “[itneeds
to be] a very cautious and well-designed, well-thought-out
approach.” The decision to design crowdsource is strategicand based
on product, people, and cost considerations. The nextquestion we
consider is how specific design attributes of theproduct concept
may guide the choice of design crowdsourcing.We searched
extensively within various literature streams forproduct design
attributes that influence product success and useracceptance (e.g.,
Poetz and Schreier 2012). We retained fiveinfluential design
attributes judged by current literature to berelevant and useful in
enhancing user response and experience,as well as two core
objectives for the utilization of external inputsfrom users.3 We
then assessed whether the crowd’s designknowledge and inputs may
help better these attributes. Next,we describe briefly how these
functional attributes influencedecisions to crowdsource (see Web
Appendix Table WA1 forreferences to these attributes from extant
literature andmanagers).
Technical complexity. We define technical complexity asthe
perceived degree of complexity due to the technical natureof the
design. New products in their initial phases are oftencomplex and
need to be simplified. The more complex thedesign, the costlier it
is to build, sell, and service a product(Radjou and Prabhu 2015)
and the greater the need for accessto a wider range of
capabilities, user involvement, and designchoices (Gann and Salter
2000; Hobday 2000). Insightsfrom theory and practice (WebAppendix
TableWA1) suggestthat managers may use design crowdsourcing to
simplify thetechnical complexity of the product. Thus, we expect
that theprobability of design crowdsourcing will increase with
in-creased levels of perceived technical complexity of the
productidea.
Usefulness. Usefulness is defined as the product’s abilityto
meet customer needs (Moldovan, Goldenberg, and Chat-topadhyay
2011). User-designed products are perceived asbetter able to meet
the needs of customers than professionallydesigned products (Poetz
and Schreier 2012). Drawing onextant literature and current
practice (Web Appendix TableWA1), we expect that the probability of
design crowdsourcingwill increase with lower levels of perceived
usefulness of theproduct idea.
3To the best of our knowledge, this is the first article to test
all fivedesign variables simultaneously in the same study. Subsets
of thesevariables are linked together theoretically in extant
literature. Forexample, Bloch (1995) groups durability, technical
sophistication,and ease of use as product-related beliefs created
or influenced byproduct form and classifies novelty as affecting
how consumersperceive the product relative to other products. Noble
and Kumar(2010) divide design elements into three categories:
rational value,kinesthetic value, and emotional (differentiating)
value. Our fiveidentified dimensions thus emphasize function
(reliability andtechnical complexity providing rational value),
user experience(ease-of-use and usefulness providing kinesthetic
value), and dif-ferentiation (novelty providing emotional
value).
Design Crowdsourcing / 111
-
Reliability. Perceived reliability relates to how well aproduct
is likely to perform, encompasses aspects such asdurability and
dependability (e.g., Grewal et al. 1998), andinfluences perceptions
of value and purchase intentions.Managers may look to the crowd for
ideas on enhanc-ing reliability. For instance, many design
proposals onCrowdspring (a design crowdsourcing website; see Table
1)use the words “durable,” and “reliable” in describing whatthey
want in a product design sketch. Even when someone isgiven a simple
product brief, evaluations of durability canbe assessed. For
example, one industrial designer workingon a simple paper sketch,
said, “whenever I am sketching, Iwant to make sure ... it looks
durable, that’s going to have tocome across in the overall design”
(Troy 2015). Thus, weexpect that the probability of design
crowdsourcing willincrease with lower levels of perceived
reliability of theproduct idea.
Usability. We define perceived usability as the expectedextent
of effort (physical or mental) required to use thenew product.
March (1994, p. 144) notes that “user-centereddesign ...
encompass[es] the cognitive aspects of using andinteracting with a
product, or how logical and natural aproduct is to use.”Thus, user
inputsmaybevaluable in enhancingusability of the concept, which can
be assessed in early stages(see additional insights in Web Appendix
Table WA1). Forexample, in the product briefs submitted to
Crowdspring,managers requested a product that “is easy to setup and
use,”“will be easy to install,” and “is unobtrusive and easy to
use.”Thus, we expect that the probability of design
crowdsourcingwill increase with lower levels of perceived usability
of theproduct idea.
Novelty. Novelty refers to the degree of newness ororiginality
of the product (e.g., Moldovan, Goldenberg, andChattopadhyay 2011;
Talke et al. 2009). Poetz and Schreier(2012) demonstrate that
user-designed products scored higheron novelty than professionally
designed products (see practiceinsights in Web Appendix Table WA1).
Thus, we expect thatproduct ideas with lower perceived novelty will
have a higherlikelihood of being crowdsourced. In summary,
H3: The probability of design crowdsourcing increases with
per-ceptions of (a) higher levels of technical complexity, (b)
lowerlevels of usefulness, (c) lower levels of reliability, (d)
lowerlevels of usability, and (e) lower levels of novelty of
theoriginal product concept.
Nonlinearity of antecedents. While the associationsproposed in
H3 relate to the initial directional nature ofthe relationships,
these relationships need not be strictlymonotonic. For example,
when creating new products,Rust, Thompson, and Hamilton (2006)
recommend offeringenough functionality for the product to not be
too simplistic,but not so much that consumers perceive the product
as beingtoo difficult to use, suggesting a nonlinear effect of
usabilityon product success. Similarly, while managers noted that
theyare more likely to use crowdsourcing as technical
complexityincreases, as one of the interviewed managers noted, in
someinstances crowdsourcing is not possible, “because you
cannotexpect the general crowd to be intelligent in terms of
your
mechanical [engineering].” In the absence of a specifictheory,
we do not propose precise directions but leave it to theempirics to
model these nonlinearities. We synthesize thesecollective insights
and theories into Figures 1 (broad con-ceptual framework) and 2
(specific design crowdsourcing–performance link).
DataEmpirical Context
We collected data on new product concepts from publiclyavailable
information from Quirky, a pioneering, community-driven NPD website
where members submit new product ideasand participate in
development efforts. Staff sorted through ideasubmissions and
selected idea(s) tomove forward. Once an ideawas selected, Quirky’s
management decided what help theywanted from the community. TheWall
Street Journal describedthe process as such:
Each week, Quirky’s staff whittles down the stream of newideas
into a dozen or so top picks that are scrutinized and votedon....
At that point, engineers and designers, working out [of]a vast red
brick warehouse in New York and three otherlocations, turn sketches
into marketable products, tapping theonline community for
suggestions about design, productnames and price points. (Simon
2014)
Quirky’s community members were promised a portionof the product
sales in exchange for their participation.Quirky’s staff chose to
ask for help in designing the productand selecting the name, logo,
or pricing, or any combinationthereof. We captured whether Quirky
asked its customers toaid in the design phase. As stated on the
website, in thedesign phase, Quirky asked its community members
to“submit sketches, images, videos, and prototypes that il-lustrate
industrial design directions for [the product]. We’lluse the top
concepts as a starting point for our final design.”
FIGURE 1Conceptual Framework for Testing
Functional Design Elements
Novelty
DesignCrowdsourcing
ProductPerformance
TechnicalComplexity
Usefulness
Reliability
Usability
Initial IdeaQuality
InventorCharacteristics
Product Cost
Non-Design Elements
112 / Journal of Marketing, March 2018
-
The staff made the final decision on design selection. Quirkyhad
no obligation to utilize any ideas from the community,and the staff
selected the phases in which to involve thecommunity. Note that
this is a similar situation faced byfirms that have a product idea
and must decide whether tofurther develop the design in-house or
crowdsource thedesign.
Description of Empirical Setting and Data
Our data set includes the 86 different products sold on
theQuirky website during our data collection period in October2014.
Quirky was very transparent with details about the NPDprocess,
design contributions, and sales, which makes thisoriginal data set
unique and valuable for addressing our researchquestions. We
gathered three key pieces of information fromthe website: First, we
retrieved the raw new product idea assubmitted by the original
community member, including thetext describing the product and, if
available, pictures or sketchessubmitted by the original inventor.
(Web Appendix FigureWA1 provides an example of the Quirky design
process.)Second, we collected information on whether Quirky
sub-sequently asked the community to help with the product
design(yes/no). The third key variable we collected from the
websitewas monthly unit sales for each product, which Quirky
pub-lished on its website.
Measures: Key Dependent, Independent, andControl Variables
New product performance. To assess new product per-formance, we
used data on total sales of new products (in-cluding sales from
both its website and retailing partners [e.g.,Amazon, Costco]).
Because not all products are released at thesame time, we used
total units sold in the first year, starting withthefirst
completemonth thatQuirky reported.Of the products inour data set,
we observed sales for a full 12-month period for 66
products (77%). For the remaining 20 products, we used asimple
three-month moving average approach to estimate thesales for the
missing months.4
Decision to design crowdsource. We collected informa-tion on
whether Quirky asked the community to help in thefunctional design
of each of its selected ideas (yes/no). Thisevent was clearly
defined by Quirky as the “design phase.” Inresponse to such
requests for help, community members sub-mitted product drawings or
sketches from which the Quirkystaff selected the best one.
Community members had a highdegree of autonomy when it came to
submitting designs. Thesubmission could be similar or different
from the original idea;all that was requiredwas that it kept to the
general essence of theproduct’s purpose. Of the products in our
data set, 22 (26%) didnot go through a design phase with community
help.
Functional attributes of design. We utilized consumerratings of
the functional design attributes because managerialcrowdsourcing
decisions will be based on consumer perceptions.As one manager
noted, “This leads to your ... question [aboutwhen one incorporates
the end user]. I believe it is key to em-pathizewith the end
user(s) throughout the entire design process.”We recruited 119
undergraduate business students at a large U.S.university to assess
raw product ideas. We took a similar blockdesign approach as in
prior research and divided the 86 productsinto 14 different
blocks,with 6–7products in each block (Kornish
FIGURE 2HowDesign Crowdsourcing Affects New Product Performance:
A Closer Look at the Link Proposed in Figure 1
DesignCrowdsourcing
(Mechanism)
Knowledge Identification Identifies new sources for andtypes of
design solutions from
the crowd
New ProductPerformance
Idea Quality
Identified link (interviews andliterature review)
Knowledge Assimilation/ Exploitation
� Increases available resourcesfor NPD (to process
moreideas)
� Improved iteration andcollaboration withusers/community in
designphase
Tested link (empirical analysis ofQuirky data set)
4We used unit sales within the first year for a few reasons:
First,most products have observed sales for one year, so it limits
ex-trapolating beyondwhat is known. Second, it focuses our analysis
onperformance soon after initial launch; this timing seems
reasonablebecause Talke et al. (2009) show that product design
affects salesmost at the beginning of a product’s life cycle. Our
results are robustto other possible methods of completing yearly
sales, such asproportional annualization (Chandrasekaran et al.
2013) or mea-suring sales at three or six months (see Table WA4 in
the WebAppendix).
Design Crowdsourcing / 113
-
andUlrich 2014), to simplify the survey andminimize
respondentfatigue. Each raw design for each final product sold on
Quirky’swebsite was randomly assigned to one of the 14 blocks.5
Each respondent viewed the product concept (pictureand
description) and was asked to evaluate each design onitems relating
to technical complexity, usefulness, reliability,usability, and
novelty. Each product was evaluated seventimes, on average, by
independent raters. Responses to eachquestion were averaged across
the respondents who evalu-ated that product. The scale questions
and their reliabilitiesappear in Table 3. To check the validity of
the model, wesuccessfully tested the construct scales using a
confirmatoryfactor analysis. The root mean square error of
approximationis .085, comparative fit index is .964, Tucker–Lewis
index is.949, and average variance extracted is .800, all of
whichshow good measurement validity. While larger samples
areusually desirable when performing confirmatory factor an-alyses,
these fit statistics provide confidence that our con-structs meet
validity assumptions. All questions were on a1–7 Likert scale,
anchored by “strongly disagree” and
“strongly agree.” We had one filtering question to filter
outthose respondents who were not paying attention (discussedin
Table 3). All construct scales were averaged across theirscale
questions to create composite construct scores.
Idea quality. Alongwith the design constructs, after viewingthe
product concept, respondents answered three questions(scale items
in Table 3) capturing the purchase intent con-struct (adapted from
Schreier, Fuchs, and Dahl 2012). Similarto prior research, we use
purchase intent as the measure ofidea quality (e.g., Kornish and
Ulrich 2014).
Control variables. Wecollected data for product category,price,
and characteristics of the idea’s inventor. Our productsfall into
five categories, as defined by Quirky: electronics (37products),
home (17 products), kitchen (23 products), and traveland health (9
products combined). Given the small number ofproducts in the travel
and health categories, we reclassified theseproducts into one of
the other three categories.
Modeling Methodology and Results
We tested our hypotheses using two different models that
areintegrated in a two-step process. First, we modeled the
factorsthat influence whether firms will crowdsource product
designsusing a binary response (probit) model. In this case,
wemodeledthe dichotomous outcome (crowdsource design: yes/no)
againstthe various ratings of design attributes of the submitted
productconcepts (e.g., usability, novelty) and other variables.
Second,
TABLE 3Product Concept Constructs and Reliabilities
Construct Construct Questionsa Reliabilityb
Technical complexity 1. The design of this product seems
highlycomplex.
.869
2. This product appears very technical.
Usefulness (adapted from Moldovan,Goldenberg, and Chattopadhyay
2011)
1. This product would be beneficial. .8032. This product
fulfills a need.
Product reliability (adapted from Grewal et al.1998)
1. The product appears be reliable. .9222. The product appears
to be of good quality.3. The product appears be dependable.
Usability (adapted from Davis 1989; see alsoVenkatesh et al.
2003)
1. I would find it easy to get this product to do whatI want it
to do.
.887
2. My interaction with this product would be clearand
understandable.
3. It would be easy to become skillful in using thisproduct.
Novelty (adapted from Dahl, Chattopadhyay, andGorn 1999)
1. This product is unique. .9302. This product is original.3.
This product is one of a kind.
Initial idea quality (measured via purchase intent;adapted from
Schreier, Fuchs, and Dahl 2012)
1. I would seriously consider purchasing thisproduct right
now.
.966
2. I would actively search for this product.3. To me, purchasing
this product in the future is
highly probable.
aOn a Likert scale measured by 1 = “strongly disagree,” and 7 =
“strongly agree.”bCronbach’s alpha or correlations if two-item
scale.Notes: To filter out respondents who were not paying
attention, we included a test question worded, “If you are paying
attention, click on ‘somewhat
agree’ in the evaluation of each product.” Any respondent who
missed multiple test questions across the products they evaluated
wereremoved from the analysis. This left us with 97 of the 119
respondents.
5We note that some of the final products originated from the
sameraw design and design phase, but because we wanted to have
aunique estimate for each final product, we allowed each
finalproduct’s raw design to be rated separately. We ensured that
each ofthese designs were inserted into different blocks. We ran
robustnesschecks to demonstrate that this does not influence the
results, whichwe describe subsequently.
114 / Journal of Marketing, March 2018
-
we examined the impact of crowdsourcing product designson market
outcomes (unit sales) using an instrumental vari-able regression
controlling for the endogenous nature of thecrowdsourcing decision
(Wooldridge 2010), since the firm self-selects which products will
be crowdsourced (i.e., the event isnot purely exogenous). The
two-step process allows us toobserve the variables that influence
the decision to crowdsourcedesign and enables us to utilize the
probit model to address theendogenous nature of the crowdsourcing
variable in the modelpredicting sales.
Wooldridge (2010) outlines the following procedure fordealing
with endogenous binary variables that utilizes thefollowing
process, where Y represents the dependent variableof interest (in
our case, unit sales), G represents the binaryendogenous variable
(design crowdsourcing), Z representsthe instruments, and X
represents the vector of controlvariables: (1) estimate a binary
choice model of dichotomousvariable G on Z and a set of controls X,
(2) obtain the fittedprobabilities of Ĝ estimated in step 1, and
(3) estimate a two-stage least squares (2SLS) instrumental
regression model,regressing Y onG andX, using Ĝ as an instrument
for G. Thisprocedure has a few notable advantages. First, it takes
thebinary property of the endogenous variable into account.Other
procedures, such as the standard 2SLS, may producebiased estimates
in finite samples. Second, note that this isdifferent than directly
inserting the fitted probabilities of theprobit in place of the
endogenous variable. Using the esti-mated probabilities in place of
the dichotomous variable in astandard ordinary least squares
requires very strict assump-tions on the error terms and the
functional form to be a validoption (Adams, Almeida, and Ferreira
2009). Third, the 2SLSprocedure is robust to misspecification in
the probit modeland provides consistent estimates with
asymptotically validerrors when using standard corrections for
heteroskedasticityin the instrumental variable estimation (Adams,
Almeida, andFerreira 2009;Wooldridge 2010, p. 939, procedure
21.1).Wepresent these two models sequentially next.
Model Specification
Model for predicting product performance. The modelused to test
the relationship between crowdsourcing the designand performance,
as measured by unit sales, can be representedby the following
functional form:
LnðUnitSalesÞi = a + b1CrowdsourceDesigni+ b2IdeaQualityi+
b3CrowdsourceDesigni· IdeaQualityi + b4HolidayLaunchi+ b5LnðPriceÞi
+ d1PCharacteristicsi+ d2PCategoryi + ei,
(1)
where Ln(UnitSales)i is the natural log of all units sold
forproduct i in the first year. CrowdsourceDesigni is a
dummyvariable that takes on a value of 1 if the product designwas
crowdsourced and 0 otherwise. Idea Qualityi is the initialidea
quality of the product and is measured via purchase intentfor the
raw concept as discussed previously (Table 3). Wecreated an
interaction term between CrowdsourceDesigni and
Idea Qualityi to test for the moderating effect of idea quality
ondesign crowdsourcing. We hypothesized this interaction to
benegative, indicating that design crowdsourcing is less
impactfulfor high-quality product ideas. HolidayLaunchi is a
dummyvariable that controls for whether the product was first
in-troduced (its first fewmonths on themarket) during the
holidayseason (November or December) to control for the
positiveproliferation effect that may come from launching the
productduring a high-volume period. Ln(Price)i is the natural log
of theselling price of the product at the time of data
collection.PCharacteristicsi represents the five design-related
constructsthat we predict will influence design crowdsourcing,
alongwiththeir squared terms. We inserted these as control
variablesbecause it is possible that these constructs will also
affect theunit sales of the product. In addition, because we
predicted thatthey will influence the decision of whether to
crowdsource thedesign, we included them as controls to assure that
the crowd-sourcing variable is capturing variance unique to
crowdsourc-ing’s effect. PCategoryi represents a vector of product
categorydummies.
Instrumenting for price. In addition, price is often con-sidered
an endogenous variable, given the simultaneous re-lationship
between price and demand. Cost is used as aninstrument for price
because it is a determinant of price butremains orthogonal to the
error term (e.g., Rossi 2014). Rossi(2014, p. 666) states that “the
idea here is that costs do not affectdemand and therefore serve to
push around price (via some sortof mark-up equation) but are
uncorrelated with the unobserveddemand shock.” The cost of raw
materials does not influencedemand because consumers are not aware
of the cost of theitems. We used the cost of raw materials as an
instrument forprice. Following Kornish and Ulrich (2014), for the
Quirkydata, we used the pictures and descriptions of the final
productsbeing sold on the website to estimate the cost of the
materialsused tomanufacture the product.We recruited
threemechanicalengineering doctoral students from the same
university, with anaverage age of 28.5 years and all with
industrywork experience.These students estimated the cost of the
raw materials used toproduce the final product in a separate task.
The three doctoralstudents first researched the current market
costs of raw ma-terials (e.g., metal, plastic, cotton) and then
used the productpictures and descriptions of the final product to
estimate the costof the rawmaterials (in dollars) used
tomanufacture the product.We averaged their cost estimates to
develop an instrument forprice and utilized the instrument in the
2SLS procedure. Thecorrelation between the natural log of price and
the natural log ofthe raw material cost is .792.
Model for predicting design crowdsourcing. Followingthe
methodology outlined previously, we specify the modelpredicting
whether an item will be crowdsourced using adiscrete choice
specification. We derive a probit model forthe design crowdsourcing
decision of the new productconcept i:
PrðCrowdsourceDesignijXÞ = FðX9b + eiÞ,(2)where F is the
standard normal cdf, and
Design Crowdsourcing / 115
-
X9b = a + b1Technicali + b2Technical2i
+ b3Usefuli + b4Useful2i + b5Reliabilityi
+ b6Reliability2i + b7Usabilityi + b8Usability
2i
+ b9Noveltyi + b10Novelty2i + d1Controlsi
+ d2Instrumentsi,
(3)
where Technicali, Usefuli, Reliabilityi, Usabilityi, and
Noveltyicorrespond to their respective construct ratings. These
con-structs were measured by the construct scores developed fromthe
surveys described previously. In addition to each linear term,we
also included a quadratic term for each of these
constructs.Controlsi represents a vector of control variables. As
noted byWooldridge (2010), the probit model should contain all
ex-ogenous control variables that are inserted in Equation
1.Instrumentsi represents a vector of variables used as
instruments(explained in detail subsequently).
Instruments and Exclusion Restrictions
So that the probit (Equation 2) results can be utilized in the
2SLSprocedure for Equation 1, the probit model must contain
ad-ditional instruments (as noted by Instrumentsi in Equation
2)that are not simultaneously listed in Equation 1. These
variablesshould influence design crowdsourcing decisions but
shouldremain unrelated to unit sales. Next, we describe the
instruments(inventor characteristics and cost variables) and
justify their useas instruments (Rossi 2014).
Social network: number of community members theinventor is
following. The managers in our interviews high-lighted that a
primary motivation behind crowdsourcing isto secure the engagement
of many people. When managerscrowdsource, they must forecast the
likelihood that there will beenough potential problem-solvers
(Afuah and Tucci 2012) toensure diverse and better solutions. Thus,
we seek a variable thatwill signal to the firm that a large number
of people are likely toparticipate in providing solutions.We
propose that the inventor’ssocial network size is a good proxy for
the likelihood that a largenumber of people will be aware of the
product and, thus, willparticipate in the crowdsourcing process.
Indeed, social con-nections of the inventor are something firms
take into account(Lohr 2015). Social networks can lead to a better
crowdsourcingprocess, due to improved reciprocity, collaboration,
feedback,and integration of ideas (Piller, Vossen, and Ihl 2012).
Therefore,we have a strong rationale that the social network of the
inventorwill matter in the decision to crowdsource design.
We look for a variable that approximates the inventor’ssocial
network and meets exclusion restrictions. The Quirkycommunity
profile allowed us to capture the number of peoplethe inventor is
“following.” The act of following forms a tie inthe networks
literature (e.g., McGee, Caverlee, and Cheng2013), where the
strength of the tie is indicated bymarkers suchas reciprocity in
following. The literature predicts that there isgreater mobility of
information and social cohesion throughweak ties than through
strong ties (Granovetter 1973), whereweak ties represent links with
distant acquaintances, such as, inthis context, following someone
on an online network. Thus,irrespective of whether the follower is
followed back, the act offollowing is a tie and all such ties form
the social network.
Furthermore, businesses monitor brand-related conversationson
social media platforms to gain access to valuable infor-mation,
influential people, and relevant conversations (Kumarand
Mirchandani 2012). Similarly, by choosing to followother
people/inventors, the inventor keeps abreast of any keydevelopments
(e.g., inventions, opinions, trends). The act offollowing is an act
of engagement/listening (e.g., Crawford2009) and not an entirely
costless act, as the inventor maychoose to follow people on a
crowdsourcing platform likeQuirky depending on his or her available
time and cognitiveresources. Thus, an inventor who is following a
large number ofpeople belongs to a larger network with the
attendant reciprocaladvantages, and his or her ideas are more
likely to benefit from abetter crowdsourcing process. We show
subsequently that thismeasure is also statistically informative.
Thus, the number ofpeople the inventor follows is an informative
and relevant in-strument thatmeets exclusion restriction
requirements because itis unlikely to directly affect sales, as
thismeasure is not salient tothe average buyer on Quirky’s
website.6 In addition,the measure contains exogenous information
driving designcrowdsourcing. The number of people the inventor
follows isthe decision of the inventor and not a result of external
forces.
Product cost. We used the cost (the same instrument usedfor
price) of the item as a second instrument. Nearly everymanager we
interviewed highlighted crowdsourcing as a way toreduce
manufacturing and production costs. Thus, because thecost of goods
sold includes the cost of raw materials and theproduction cost of
the individual item, if the cost of rawmaterialsis high, managers
are more likely to look for ways to reduceproduction costs;
thiswill enable thefirm to keep the total cost aslow as possible.
As noted previously, we used the doctoralstudents to estimate the
cost of thematerials used tomanufacturethe product. This measure of
cost is a valid instrument, as it isunlikely that cost directly
affects consumer demand (Rossi2014), but it does affect the
crowdsourcing decision. We furtherincluded a quadratic term for
cost, because not all increases incostswill be associatedwith the
same increases in crowdsourcing.For example, the change in
probability of crowdsourcingbetweenitems that cost $10 versus $20
might be quite different thanbetween items that cost $200 versus
$210.
Finally, we also included interaction terms between
theinstruments—namely, cost and the number of communitymembers that
the inventor is following. The desire to lowercosts will trump
other external cues (such as the inventor’snetwork) to determine
whether the firm should crowdsource.Thus, the hypothesized
direction for the number of people theinventor is following will
hold at low levels of cost, but theeffect should be nonlinear per
our managerial insights, whichwe capture with the interaction with
cost. The probit modelindicates that all of the instruments are
significant (see Table 4),and the pseudoR2 increases from .286 to
.476with the inclusionof these instruments.We next discuss results
for Equation 2 andthen Equation 1.
6To find information about who an inventor follows, a
consumerwould have to consciously click through the website to find
theinventor’s profile. Furthermore, much of Quirky’s sales occurs
inother retailer settings, for example, in Walmart, where such
retailcustomers will know nothing about individual inventors.
116 / Journal of Marketing, March 2018
-
Characteristics That Influence DesignCrowdsourcing
The summary statistics and correlation matrix are includedin
Table WA2 of the Web Appendix.7 Table 4 displays theresults for
Equation 2. Heteroskedasticity-robust standarderrors are used in
computing the Wald tests. The results showthat while the estimated
coefficients for Useful and Noveltyare not significantly different
from zero, the linear andquadratic terms for all the other
constructs are statisticallysignificant. Specifically, Usabilityi
has a negative linear term(b = -62.245, p = .001) and a positive
quadratic term (b =5.418, p = .001); this suggests that, initially,
the probability ofcrowdsourcing design decreases with an increase
in usability.After the perceived usability reaches a certain level,
theprobability of crowdsourcing design increases. This suggeststhat
firms are more likely to crowdsource designs that appearoverly
difficult or too easy to use. The lowest probabilityoccurs at about
its mean, where the squared term dominates thelinear term, around
5.74 on the 7-point Likert scale (» 62.245/[2· 5.418]). The effect
of perceived reliability of the raw concepton the choice to
crowdsource design follows a somewhatsimilar pattern. The negative
linear term (b = -32.391,p = .004) suggests that as the perceived
reliability of theconcept increases, the probability of
crowdsourcing designdecreases. Thus, at low levels of perceived
reliability, the firmis more likely to seek the help of the
community in developingthe design. The positive quadratic term (b =
3.158, p = .005)suggests that after a certain level of
reliability—which occursat roughly 5.13 (» 32.391/[2 · 3.158]),
increases in perceivedreliability are not associated with a
decrease in designcrowdsourcing.
Technical complexity (Technicali) of the product followsa
pattern opposite those of usability and reliability. The
positivelinear coefficient (b = 4.322, p = .006) demonstrates that
themore technical a product idea is, the more likely the firm is
tocrowdsource design. However, with higher levels of
technicalcomplexity, marginal increases in technical complexity
areassociated with a decreasing probability of crowdsourcing,
asindicated by its negative quadratic term (b = -.685, p =
.001).The inverted U-shape of technicality shows that the
highestprobability of crowdsourcing is 3.155 (» 4.322/[2 · .685])
onthe 7-point scale, with the lowest probability occurring at
theends. This supports the notion presented by some of
themanagerial insights that firms are less likely to
crowdsourcedesigns that are too technical, because the community
will lackthe needed expertise, and supports extant research that
suggeststhat crowdsourcing is less likely when a firm doubts the
crowd’s
ability/expertise to evaluate solutions (Afuah and
Tucci2012).
Overall, these results suggest a strong relationship be-tween
the probability of crowdsourcing a design and the designattributes
of the raw product concept. The constructs we hy-pothesized, with
the exception of usefulness and novelty, wererelated to the
decision to design crowdsource, in support of H3a,H3c, and H3d. We
discuss further validation of the importance ofthese design
characteristics on design crowdsourcing decisionsnext.
Validating the Importance of Design Characteristicson Design
Crowdsourcing Decisions
We next demonstrate that these design constructs
influencesimilar decisions in a different data context. The
followinganalysis is not meant to replicate the exact same decision
but toprovide convergent evidence that design constructs drive
designcrowdsourcing decisions.
Our second data set comes from Crowdspring, an
onlinecrowdsourcing platform where firms post design
challenges.Crowdspring allows clients in need of design help to
posttheir requirements and get responses from the crowd (seeWeb
Appendix Figure WA2). On average, a brief receivesover 90 entries,
and the client typically picks a winningdesign from these entries.
We obtained data on 27 completeddesign projects from Crowdspring.
We retained consumer-oriented products and deleted fashion-related
products (e.g.,clothes, jewelry) because these projects deal with
aestheticsmore than functional design, leaving a total of 20
projects.We collected the winning design(s) and selected ten
other“nonwinning” designs at random for each of the designbriefs.
The submitted product designs (the winning designsplus the ten
chosen at random) were evaluated using the sameconstruct scales
(Table 3) by respondents from Amazon’sMechanical Turk (MTurk). Each
respondent (there was anaverage of eight respondents per product)
saw a picture of thedesign submission and a write-up describing the
product.Their answers were averaged to form the constructs’
scores;all construct reliabilities (Cronbach’s alpha) or
correlations(for two-item measures) were greater than .80.
Using this new data set, we tested whether the samedesign
characteristics that affected the earlier designcrowdsourcing
decision also influence which design man-agers select as the
winning design from all submissions (andpresumably choose to
implement). We utilized a binaryregression method similar to the
previous analysis becausethe outcome variable is dichotomous
(whether the designwas chosen or not), with one modification: we
controlled forthe fact that each design is not independent but is
clusteredwithin a specific project and that the winner is
determinedfrom the specific cluster. We did this using a
conditionallogistic regression model, which is fitted to such
situationswhere the data are based on matched cases (groups),
viathe “clogit” command in STATA 13. This controls forthe
conditional nature of the outcome variable where thelikelihood
estimation is calculated relative to the group.We included all five
design characteristics from Equation 2(for the results, see Table
WA3 of the Web Appendix).
7There may be a possible concern about the quadratic termshaving
a multicollinearity problem as a result of their linear terms.As
Allison (2012) notes, high correlation between variables andtheir
product is expected and “is not something to be concernedabout,
because the p-value for [the product] is not affected by
themulticollinearity,... so the multicollinearity has no adverse
conse-quences.” Furthermore, the variance inflation factor analysis
of thebase models for Equation 2 (without interactions and
nonlinearterms) confirms that multicollinearity is not an issue due
to highcorrelations between constructs, because all of the variance
inflationfactor values were substantially less than ten.
Design Crowdsourcing / 117
-
Two of the significant constructs from our prior analysis
onQuirky data are also significant in this data set (p <
.10)—usability and reliability—with both linear and quadratic
termssignificant. The probability of a design being chosen (as
awinner) increases as Usability increases (b = 10.612, p =
.079);its quadratic term, Usability2 (b = -1.004, p = .069),
indicates atapering effect, suggesting that it increases at a
decreasing rate.Reliability positively increases the probability of
a design beingchosen, (b = 18.877, p = .091), with Reliability2
indicating atapering effect (b = -1.882, p = .081). Technical
complexitywas not statistically different from zero for either the
linearor quadratic terms (p > .10). It could be that there may
be ahigher degree of variation in this construct when measur-ing
across different products (such as with Quirky), butnot across
different designs for the same product, as inCrowdspring, where
managers may have been more explicitabout technicality. The
significant results for two designconstructs, usability and
reliability, and their quadratic terms,across data contexts provide
convergent evidence for theirinfluence on crowdsourcing
decisions.
Design Crowdsourcing and New ProductPerformance
Wenext present the results on the impact of design
crowdsourcingon postlaunch new product performance using data from
Quirky.
Table WA2 of the Web Appendix presents the summarystatistics and
correlation matrix. Equation 1 is estimated with2SLS using the
instrumental procedure, as previously noted,
where the predicted result from the probit model (Equation 2)
isused as an instrument. Table 5 shows the results of Equation
1utilizing two nested models. First, we present the results of
themain-effects-only model (excluding the multiplicative term
b3)and then present the results of the full model with
interactioneffects. Both models use heteroskedasticity-robust
standarderrors. Table 5 also shows the first-stage F-statistics and
thepartial R-squares of the instruments as estimated in the
first-stage regressions of the 2SLS procedure. The diagnostics
forthe crowdsourcing dummy, the crowdsourcing · idea
qualityinteraction, and price instruments show that, collectively,
ourinstruments are not weak (Stock and Watson 2003).
The results in Table 5 demonstrate an interesting
relationshipbetween design crowdsourcing and unit sales. Model 1
(maineffects only) shows that the design crowdsourcing dummy
doesnot significantly affect sales (b = .566, p = .527). This
indicatesthat the effect of design crowdsourcing, on average, is
not sta-tistically different from zero. However, Model 2, the full
modelthat includes interaction, presents a more nuanced picture.
Theestimate for CrowdsourceDesign (b = 12.824, p = .002) andthe
interaction between CrowdsourceDesign and IdeaQuality(b = -3.188,p=
.001) are both statistically significant.We remindthe reader that
the beta coefficients in Model 1 (the main-effects-only model)
represent the estimation of the main effect, or theaverage effect
across the dependent variable based on the con-ditional mean
function E(y|x) (Baum 2013). The coefficient forCrowdsourceDesign
in Model 2 (the full model with interaction)represents the
estimation of the simple effect or the estimatedimpact of
CrowdsourceDesign when IdeaQuality is at zero (for a
TABLE 4Probit Model: Predicting Design Crowdsource
Coefficient Est. SE p-Value
Functional Design ElementsTechnical 4.322 1.568 .006Technical2
-.685 .206 .001Useful -3.698 12.283 .763Useful2 .421 1.155
.715Reliability -32.391 11.311 .004Reliability2 3.158 1.133
.005Usability -62.245 19.018 .001Usability2 5.418 1.663 .001Novelty
4.509 4.886 .356Novelty2 -.543 .541 .316
Instruments (Nondesign Elements)Social Network: Following .003
.001 .004Ln(ItemCost) 1.808 .957 .059Ln(ItemCost)2 -.530 .239
.027Following · Ln(ItemCost) -.009 .003 .001Following ·
Ln(ItemCost)2 .006 .002 .000
Additional Controls from Equation 1Idea quality 1.534 .507
.002Holiday launch 1.204 .495 .015Category dummies included
YesObservations (N) 86Log pseudolikelihood -25.638Pseudo R2
.476
Notes: For brevity, product category dummies along with the
constant are estimated but not displayed. Robust standard errors
are presented. Thez-values for the instruments are as follows:
Following (z = 2.91), Ln(ItemCost) (z = 1.89), Ln(ItemCost)2 (z =
-2.22), Following · Ln(ItemCost)(z = -3.39), Following ·
Ln(ItemCost)2 (z = 3.62).
118 / Journal of Marketing, March 2018
-
discussion of simple effects in interactivemodels, see
Echambadiand Hess 2007). The significance of CrowdsourceDesign in
themodel with interactions and lack of significance in the
modelwithout interactions suggest that crowdsourcing the design
helpssales, but onlywhen the concept has low IdeaQuality. The sign
ofthe interaction term helps make sense of this distinction.
Thenegative interaction term suggests that the positive effect
ofdesign crowdsourcing on sales dissipates as the idea quality of
theproduct concept increases.
To increase the managerial relevance of our findings, weaim to
show that crowdsourcing the design helps, on average, allproducts
with low levels of IdeaQuality (not only those at zero).We measure
the marginal effect at different low levels ofIdeaQuality. The
marginal effect of CrowdsourceDesign ispositive and significant at
various levels of IdeaQuality. In fact,the positive effects do not
become insignificant (p > .10) untilaround IdeaQuality’s mean.
Thus, we find support for H1 whenthe product idea shows room for
improvement, and we findsupport for H2. Table WA4 in the Web
Appendix shows ro-bustness of this analysis to alternativemeasures
of sales.We alsoshow that the results are robust, accounting for
the fact that someproducts originated from the same raw design by
using clusteredstandard errors (Tables WA5 and WA6 in the Web
Appendix).
What Product Functionalities Does DesignCrowdsourcing
Influence?
A logical follow-up question to the preceding analysis iswhether
design crowdsourcing improves the functional designattributes from
idea to final product. To test this notion, wecollected additional
data to assess the design of the final productas presented on the
Quirky website. We replicate the main-effects-only model (Model 1
from Table 5), replacing unit sales(the previous dependent
variable) with the change scores of thethree design characteristics
found to be significant in the probitmodel (reliability, technical
complexity, and usability) as thenew dependent variables. We assess
the change scores, usingconsumer ratings, bymeasuring the
improvement from the initialproduct idea to the final product for
each of the measured designcharacteristics. We utilized the same
design, same filteringquestions, and the same construct questions
used for the initialproduct ideas to assess the design
characteristics of the finalproduct, using students from the same
population, but ex-cluding any respondentswho participated in
evaluating the rawconcepts. After aggregating the questions into
the final con-structs, we calculated a change score for each of the
designconstructs by subtracting the initial rating from the final
rating.
TABLE 5The Effect of Crowdsourcing the Design on Unit Sales
A: Results for Equation 1
Model 1: Main-Effects-Only Model Model 2: Full Model
Coefficient Est. SE p-Value Coefficient Est. SE p-Value
CrowdsourceDesign .566 .895 .527 12.824 4.116 .002IdeaQuality
-.149 .431 .729 1.512 .656 .021CrowdsourceDesign · IdeaQuality
-3.188 .945 .001ControlsLn (Price) -.965 .373 .010 -1.627 .482
.001Technical -1.517 1.115 .173 -2.641 1.438 .066Technical2 .198
.148 .180 .337 .192 .079Useful -1.749 4.503 .698 -.627 5.080
.902Useful2 .138 .454 .761 .091 .495 .854Reliability -3.144 1.627
.053 .047 2.806 .987Reliability2 .407 .173 .019 .192 .264
.467Usability 2.320 3.829 .545 4.826 4.622 .296Usability2 -.251
.372 .499 -.530 .456 .245Novelty -3.846 3.265 .239 -5.703 3.437
.097Novelty2 .458 .357 .200 .702 .376 .062HolidayLaunch .047 .354
.893 -.118 .440 .788Category dummies included Yes YesObservations
(N) 86 86R-squareda .295 .208
B: Instrument Diagnostics for Equation 1
Model 1: Main-Effects-Only Model Model 2: Full Model
First-Stage F-Stat. Partial R2 First-Stage F-Stat Partial R2
CrowdsourceDesign 29.618 .377 19.905 .379CrowdsourceDesign ·
IdeaQuality 28.056 .427Ln(Price) 37.569 .519 28.952 .520
aR-squared is shown for directional purposes only. R-squared for
2SLSmodels is not interpreted the same as ordinary least squares
(percentage ofvariance explained). See Wooldridge (1999).Notes:
Robust standard errors presented. A constant and category dummies
are estimated but not displayed for brevity.
Design Crowdsourcing / 119
-
For example, DReliabilityi, the dependent variable, wouldbe
calculated as Reliablityi,final – Reliabilityi,initial,
whereReliabilityi,final relates to the score for perceived
reliability ofthe final product and Reliabilityi,initial relates to
the score forperceived reliability of the initial idea. This model
used thesame instrumental variables procedure as before and
controlsfor the initial level of the product characteristic
ratings.
Three different models were run, using
DReliabilityi,DTechnicali, and DUsabilityi as the dependent
variables, re-spectively, and design crowdsourcing dummy as the key
in-dependent variable. The results (Web Appendix Table WA7)show
that the design crowdsourcing dummy positively in-fluences
DReliabilityi (b = .548, p = .079) and DUsabilityi(b = .772, p =
.000), but not DTechnicali (b = -.211, p = .542).These results
suggest that design crowdsourcing enhances per-ceived reliability
andusability fromproduct idea tofinal product.8
Data Robustness Checks
Use of student sample. Another potential concern withour data
could be that we used undergraduate businessstudents to assess
design ratings and idea quality. Theprimary objective of
crowdsourcing is to design productsthat are aligned with what
people want. Thus, it is clear thatconsumer preferences around
product design attributesdrive the crowdsourcing decision. Our use
of student sur-veys is representative of these considerations of
obtainingconsumer preferences to aid managerial decision
making.Students are often used as proxies for general
consumers(e.g., Aaker and Keller 1990; Larson and Billeter
2013).Furthermore, students are a specific target segment
forQuirky, as evident from some of the product descriptions,such as
“dorm occupants needn’t schlep their shower shoes;just hook a cord
around them and they’re along for the ride.”Therefore, students are
representative of a strong consumerbase for Quirky. However, to
show that the student ratingsare similar to the ratings from other
general segments, wecollected product ratings for a subsample of
the sameproducts from two different groups of respondents: onegroup
recruited from MTurk and another group of businessprofessionals
(master of business administration [MBA]graduates, using a panel
provided by Qualtrics).9 We ran-domly selected 42 products (3
products from each of the 14blocks) and had each group rate the
products on the same
construct questions shown in Table 3. We compared theseratings
with the earlier ratings for the same 42 productstaken from our
original (undergraduate business) sample.We first performed a
comparison of between-sample sim-ilarities using correlations
between the samples, and then acomparison of within-sample
similarities using the Jennrichtest (Jennrich 1970).
The high and significant correlation between theMTurk ratings
and the undergraduate ratings across the42 products for technical
complexity (.714, p < .001),usefulness (.528, p = .003),
reliability (.305, p = .049),usability (.609, p < .001), and
novelty (.449, p = .003)demonstrates that the construct scores are
consistentacross these different respondent groups. Next, we
usedthe Jennrich test of equality of correlation matrices,
whichformally tests whether the correlational structure for thefive
constructs differs between the groups (Jennrich1970). In other
words, it tests whether the correlationmatrix for the student
sample is similar to the correlationmatrix in the MTurk sample
(e.g., Compas et al. 1989;Gande and Parsley 2005), without
requiring the assumption ofequal means or standard deviations
(Gande and Parsley 2005).The test shows no significant difference
(p = .87). In otherwords,we fail to reject the null hypothesis that
the correlationalstructures are equal, indicating that the
relationships amongconstructs are similar across samples.
Similarly, if we includethe design crowdsourcing dummy and the idea
quality rating inthe correlation matrices, the test again shows no
significantdifference (p = .61).
Next, we recruited 135 business professionals using aQualtrics
panel to rate the same 42 products. All constructsbetween the two
samples (students and MBA pro-fessionals) were significantly
correlated (p < .01), exceptfor Reliability, demonstrating that
the ratings on theconstruct scores are generally consistent across
these tworespondent groups. The Jennrich test showed no
signifi-cant difference (p = .59) between matrix
structures.Similarly, if we include the design crowdsourcing
dummyand rating on idea quality, the test again shows no
sig-nificant difference (p = .72). We elaborate on the Jennrichtest
and other robustness checks regarding our sample inthe section on
additional robustness checks in the WebAppendix.
Nonlinearities. One possible question is whether theuse of all
of the quadratic terms is warranted in the modelpredicting the
probability of design crowdsourcing. Wehave noted that extant
theory suggests that at least two ofthe constructs should be
nonlinear (technical complexityand usability). Therefore, we
replicate Equation 2 butinclude only two quadratic terms for
technical complexityand usability (see Web Appendix Table WA8). The
sameconstructs that are significant in the previous analysis(Table
4) are significant (Technical, Technical2, Re-liability, Usability,
and Usability2) and in the same di-rection. In addition, the same
constructs that werepreviously not significant (Useful and Novelty)
are stillnot significant.
8We rationalize post hoc that technical complexity may be
moresignificant in the antecedents model than in the
change-scoreanalysis because technical complexity may be more
internal and,thus, significant when firm capabilities matter in the
decision ofwhether to crowdsource. The change-score results
demonstrate thatcrowdsourcing improves those attributes of design
that are perhapsmore user-centric.
9For the MTurk sample, we recruited 165 respondents to rate
42products (average age = 39.8 years old, 39% male, medianhousehold
income: $50,000–$99,999). Respondents in the Qualtricssample had
all graduated from an MBA program (average age = 48years old,
64%male, medium household income: $50,000–$99,999,average work
experience in business: 18 years). The same scaleswere used for the
five design constructs and the average rating acrossthe respondents
(at least five respondents per product) was obtained.All Cronbach’s
alpha/correlations are above .80 for the MTurksample and above .70
for the Qualtrics sample.
120 / Journal of Marketing, March 2018
-
DiscussionThe use of design crowdsourcing to seek external
inputs duringdesign is emerging as a significant practice. Our
article is one ofthe first to build and test a theoretically
grounded model offactors that influence the design crowdsourcing
decision and theeffect of design crowdsourcing on performance,
providingimplications for both academics and managers.
Contributions to Theory
Knowledge management theory. Our exploratory in-sights
contribute to the knowledge management literature byrevealing how
design crowdsourcing, as a mechanism, canhelp improve the NPD
process and performance. Designcrowdsourcing aids in knowledge
identification by aggre-gating diverse sources of user-based design
knowledge andextracting novel, workable, andmeaningful design
solutions.The literature on ideation crowdsourcing reveals
similarinsights on the use of the crowd as a resource base for
ideasduring the opportunity identification stage. We complementthis
research by illustrating how crowdsourcing can strate-gically tap
into the crowd during the critically importantdesign stage. Design
crowdsourcing bolsters opportunityexploitation by supplementing NPD
resources and creating amore collaborative process of integrating
external solutionswith in-house guidance, thereby contributing to
productdevelopment and performance.
Crowdsourcing theory. We add to the extant crowd-sourcing
literature by illuminating the antecedents of designcrowdsourcing
and by examining design crowdsourcing’seffect on new product
performance. Exploratory inter-views and results reveal that the
inherently iterative, user-driven, and evolutionary process of
design crowdsourcingcan lead to a more focused search for
innovative solutionswhile simultaneously enhancing product
effectiveness. Ouranalysis reveals that product ideas with
significant need forimprovement may likely benefit the most from
this iterativeprocess that allows for crowd-driven refinement and
en-hancement of such ideas.
Product design theory. We contribute to the literature onproduct
design by finding that design elements—usability,reliability, and
technical complexity—matter in influenc-ing crowdsourcing
decisions. Surprisingly, usefulness andnovelty do not emerge as
significant drivers from our data.One post hoc explanation is that
novelty and usefulnessmatter during idea selection, where the
emphasis may be ondifferentiation, while the other dimensions are
of conse-quence during design selection, where the emphasis may
beon objective functionality and user experience. Based on
thedesign literature (Noble and Kumar 2010), it seems thatmanagers
may be seeking utility from design crowdsourcingto enhance function
(rational value), and user experience(kinesthetic value), rather
than differentiation (emotionalvalue). Furthermore, this research
establishes that the crowdcan serve as a knowledge resource for
design solutions, anddesign crowdsourcing can improve user
perceptions ofdesign from ideas to products.
Managerial Implications
Our results provide three important managerial implica-tions.
First, whereas managers may fear losing control of thedesign
process by opening it up, our interviews indicate thatthey can
maintain better control over the design process,while creating
slack for their research and development/design team, through the
process of engagement/iterationwith users/designers by selecting
appropriate designcrowdsourcing platforms. Furthermore, managers
are facedwith pressure to generate greater numbers of
innovativeproducts while being constrained by internal
resourcelimitations. Therefore, they often prioritize only their
bestproduct ideas and concepts, discarding many others. Ourresults
suggest that design crowdsourcing can help managersmove a greater
number of ideas through development by usingthe community’s help in
making (initially) less-promising ideasmarketable. Thus, we address
the question we posed previously:there is incremental value to be
extracted from even initially less-promising ideas. Rather than
discard such ideas, firms may useexternal sources of knowledge to
develop them, and interactwiththese external sources extensively to
ensure that the outcome isof high quality. Newer crowdsourcing
firms, such as Crowd-spring, are now setting up systems for the
idea generator (clientfirm) to provide feedback to the community as
the communityaids in the design process. This feedback process may
be ratedand monitored by the crowdsourcing platform. It is this
in-teractive and iterative process of design and development
thateventually moves ideas into production, and herein lies the
truevalue of design crowdsourcing.
Second, our analysis suggests design crowdsourcing in-creases
the perceived reliability and usability from ideation tofinal
product. Managers of client firms aiming to improvespecific
functional attributes of design may turn to crowd-sourcing as a
supplementary design resource.
Third, we provide insights to the crowdsourcing platforms,as
well as the client firms, on better managing the process
ofcrowdsourcing. First, design crowdsourcing firms are
in-creasingly facing pressure from members of design commu-nities,
who perceive a threat posed by the availability ofthousands of
low-cost designs provided by the crowd. (Grefe2016). Our research
suggests that design crowdsourcing canhelp improve specific design
functionalities through a processof iteration and feedback. Design
crowdsourcing firms can (re)position themselves as intermediaries
that help solve genuineproduct needs. Second, this research
emphasizes the iterativeprocess of design. However, given the
start-up nature of manyof the crowdsourcing platforms, there may be
di